Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| import re | |
| import html | |
| from transformers import pipeline | |
| import numpy as np | |
| # Initialize pipelines | |
| ocr = pipeline("image-to-text", model="microsoft/trocr-base-printed") # Consider upgrading to trocr-large-printed | |
| classifier = pipeline("zero-shot-classification", model="facebook/bart-large-mnli") # Consider fine-tuning for phishing | |
| # Common phishing indicators | |
| SUSPICIOUS_PHRASES = [ | |
| "urgent", "immediately", "password", "account locked", "wire transfer", | |
| "bank verification", "click here", "verification code", "credit card", | |
| "suspended", "login now", "reset your password", "act now", "unusual activity", | |
| "security alert", "confirm your identity", "gift card", "lottery winner" | |
| ] | |
| # URL detection regex | |
| URL_PATTERN = r'http[s]?://(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\\(\\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+' | |
| def extract_text_from_image(image): | |
| if image is None: | |
| return "" | |
| try: | |
| result = ocr(image) | |
| return result[0]["generated_text"] if result else "" | |
| except Exception as e: | |
| return f"Error processing image: {str(e)}" | |
| def analyze_text(text): | |
| if not text.strip(): | |
| return "", "", gr.update(visible=False) | |
| # Zero-shot Classification | |
| candidate_labels = ["Phishing Email", "Legitimate Email"] | |
| result = classifier(text, candidate_labels=candidate_labels) | |
| label = result["labels"][0] | |
| confidence = result["scores"][0] | |
| # Determine risk level and styling | |
| if label == "Phishing Email": | |
| if confidence > 0.8: | |
| alert_html = """ | |
| <div style="padding: 20px; background: linear-gradient(to right, #ffebee, #ffcdd2); | |
| border-radius: 12px; box-shadow: 0 2px 4px rgba(0,0,0,0.1); margin-bottom: 20px;"> | |
| <div style="display: flex; align-items: center; gap: 12px;"> | |
| <span style="font-size: 24px;">⚠️</span> | |
| <h3 style="color: #c62828; margin: 0; font-size: 18px;">High Risk Detected - Likely Phishing Attempt</h3> | |
| </div> | |
| </div> | |
| """ | |
| else: | |
| alert_html = """ | |
| <div style="padding: 20px; background: linear-gradient(to right, #fff3e0, #ffe0b2); | |
| border-radius: 12px; box-shadow: 0 2px 4px rgba(0,0,0,0.1); margin-bottom: 20px;"> | |
| <div style="display: flex; align-items: center; gap: 12px;"> | |
| <span style="font-size: 24px;">⚡</span> | |
| <h3 style="color: #ef6c00; margin: 0; font-size: 18px;">Medium Risk - Suspicious Content Detected</h3> | |
| </div> | |
| </div> | |
| """ | |
| else: | |
| alert_html = """ | |
| <div style="padding: 20px; background: linear-gradient(to right, #e8f5e9, #c8e6c9); | |
| border-radius: 12px; box-shadow: 0 2px 4px rgba(0,0,0,0.1); margin-bottom: 20px;"> | |
| <div style="display: flex; align-items: center; gap: 12px;"> | |
| <span style="font-size: 24px;">✅</span> | |
| <h3 style="color: #2e7d32; margin: 0; font-size: 18px;">Low Risk - Likely Legitimate</h3> | |
| </div> | |
| </div> | |
| """ | |
| # Find suspicious phrases and URLs | |
| found_phrases = [] | |
| text_lower = text.lower() | |
| for phrase in SUSPICIOUS_PHRASES: | |
| if phrase in text_lower: | |
| found_phrases.append(phrase) | |
| urls = re.findall(URL_PATTERN, text) | |
| # Generate detailed analysis report with modern styling | |
| report = [ | |
| "<div style='background: white; padding: 24px; border-radius: 12px; box-shadow: 0 2px 8px rgba(0,0,0,0.05);'>", | |
| "<h3 style='color: #1a237e; margin-top: 0;'>Analysis Details</h3>", | |
| f"<div style='display: flex; gap: 20px; margin-bottom: 20px;'>", | |
| f"<div style='flex: 1; padding: 16px; background: #f5f5f5; border-radius: 8px;'>", | |
| f"<strong>Confidence Score:</strong> {confidence:.1%}", | |
| "</div>", | |
| f"<div style='flex: 1; padding: 16px; background: #f5f5f5; border-radius: 8px;'>", | |
| f"<strong>Classification:</strong> {label}", | |
| "</div>", | |
| "</div>" | |
| ] | |
| if found_phrases or urls: | |
| report.append("<div style='margin-top: 20px;'>") | |
| if found_phrases: | |
| report.extend([ | |
| "<h4 style='color: #d32f2f;'>🚩 Suspicious Elements Detected:</h4>", | |
| "<ul style='list-style-type: none; padding-left: 0;'>" | |
| ]) | |
| for phrase in found_phrases: | |
| report.append(f"<li style='margin-bottom: 8px; padding: 8px 12px; background: #ffebee; border-radius: 6px;'>Found: '{html.escape(phrase)}'</li>") | |
| report.append("</ul>") | |
| if urls: | |
| report.extend([ | |
| "<h4 style='color: #d32f2f;'>🔗 Detected URLs:</h4>", | |
| "<ul style='list-style-type: none; padding-left: 0;'>" | |
| ]) | |
| for url in urls: | |
| report.append(f"<li style='margin-bottom: 8px; padding: 8px 12px; background: #e3f2fd; border-radius: 6px; word-break: break-all;'>{html.escape(url)}</li>") | |
| report.append("</ul>") | |
| report.append("</div>") | |
| else: | |
| report.append("<p style='color: #2e7d32;'>✅ No common suspicious phrases or URLs detected.</p>") | |
| report.append("<div style='margin-top: 20px; padding: 16px; background: #e3f2fd; border-radius: 8px;'>") | |
| if confidence > 0.9: | |
| report.append("<p style='margin: 0;'><strong>🔴 High confidence in classification - exercise extreme caution!</strong></p>") | |
| elif confidence > 0.7: | |
| report.append("<p style='margin: 0;'><strong>🟡 Moderate confidence - review carefully and verify sender.</strong></p>") | |
| else: | |
| report.append("<p style='margin: 0;'><strong>🟢 Low confidence - but always remain vigilant.</strong></p>") | |
| report.append("</div></div>") | |
| return alert_html, "\n".join(report), gr.update(visible=True) | |
| def process_input(text_input, image_input): | |
| if text_input.strip(): | |
| return analyze_text(text_input) | |
| if image_input is not None: | |
| extracted_text = extract_text_from_image(image_input) | |
| if extracted_text.strip(): | |
| return analyze_text(extracted_text) | |
| return ( | |
| """<div style="padding: 20px; background: #fff3e0; border-radius: 12px; margin-bottom: 20px;"> | |
| <h3 style="color: #ef6c00; margin: 0;">⚠️ OCR Processing Error</h3> | |
| </div>""", | |
| "Could not extract text from image. Please ensure the image contains clear, readable text.", | |
| gr.update(visible=False) | |
| ) | |
| return "", "Please provide either text or an image to analyze.", gr.update(visible=False) | |
| # Custom theme | |
| custom_theme = gr.themes.Soft().set( | |
| body_background_fill="#f8f9fa", | |
| block_background_fill="white", | |
| block_label_background_fill="*background-3", | |
| input_background_fill="white", | |
| button_primary_background_fill="#1a237e", | |
| button_primary_text_color="white", | |
| button_secondary_background_fill="#e0e0e0", | |
| button_secondary_text_color="#333", | |
| ) | |
| # CSS for responsiveness and polish | |
| custom_css = """ | |
| .container { max-width: 1000px; margin: auto; } | |
| .header { text-align: center; margin-bottom: 2rem; } | |
| .tool-description { max-width: 800px; margin: 0 auto 2rem auto; } | |
| .input-section { margin-bottom: 2rem; } | |
| .analysis-section { margin-top: 2rem; } | |
| @media (max-width: 600px) { | |
| .header h1 { font-size: 1.8rem; } | |
| .input-section { padding: 0 10px; } | |
| .gr-button { width: 100%; margin-bottom: 10px; } | |
| } | |
| """ | |
| # Create Gradio interface | |
| with gr.Blocks(theme=custom_theme, css=custom_css) as demo: | |
| gr.HTML(""" | |
| <div class="header"> | |
| <h1 style="color: #1a237e; font-size: 2.5rem; margin-bottom: 1rem;">🛡️ AI Phishing Guard</h1> | |
| <p style="color: #555; font-size: 1.2rem;">Protect yourself from phishing attempts with AI-powered analysis</p> | |
| </div> | |
| """) | |
| with gr.Row(equal_height=True): | |
| with gr.Column(): | |
| gr.HTML(""" | |
| <div class="tool-description"> | |
| <h3 style="color: #1a237e;">How to Use</h3> | |
| <ol style="color: #555; line-height: 1.6;"> | |
| <li>Paste message text or upload a screenshot</li> | |
| <li>Analyze instantly as you type or click 'Analyze' for images</li> | |
| <li>Review the detailed results and stay cautious</li> | |
| </ol> | |
| <div style="background: #e3f2fd; padding: 16px; border-radius: 8px; margin-top: 1rem;"> | |
| <h4 style="color: #1a237e; margin-top: 0;">This tool detects:</h4> | |
| <ul style="color: #555; margin-bottom: 0;"> | |
| <li>Suspicious language patterns</li> | |
| <li>Common phishing phrases</li> | |
| <li>URLs and potential threats</li> | |
| <li>Urgency indicators</li> | |
| </ul> | |
| </div> | |
| </div> | |
| """) | |
| with gr.Tabs(): | |
| with gr.TabItem("✏️ Text Input"): | |
| text_input = gr.Textbox( | |
| lines=8, | |
| label="Message Text", | |
| placeholder="Paste email or message content here...", | |
| elem_id="text_input" | |
| ) | |
| with gr.TabItem("📷 Screenshot Upload"): | |
| image_input = gr.Image( | |
| label="Upload Screenshot", | |
| type="pil", | |
| elem_id="image_input" | |
| ) | |
| with gr.Row(): | |
| analyze_button = gr.Button("🔍 Analyze", variant="primary", size="lg") | |
| clear_button = gr.Button("🗑️ Clear", variant="secondary", size="lg") | |
| loading = gr.HTML("<div style='text-align: center; padding: 10px;'>Analyzing... <span>⏳</span></div>", visible=False) | |
| with gr.Column(visible=False) as output_col: | |
| alert_html = gr.HTML() | |
| analysis = gr.HTML() | |
| # Examples | |
| examples = [ | |
| ["Subject: URGENT - Account Security Alert\n\nDear User,\n\nWe detected unusual activity in your account. Click here immediately to verify your identity and reset your password at https://fakebank.com/reset. If you don't respond within 24 hours, your account will be suspended.\n\nBank Security Team", None], | |
| ["Subject: Team Meeting Tomorrow\n\nHi everyone,\n\nJust a reminder that we have our weekly team meeting tomorrow at 10 AM in the main conference room. Please bring your project updates.\n\nBest regards,\nSarah", None], | |
| ] | |
| gr.Examples( | |
| examples=examples, | |
| inputs=[text_input, image_input], | |
| outputs=[alert_html, analysis, output_col], | |
| fn=process_input, | |
| cache_examples=True | |
| ) | |
| # Event handlers | |
| def with_loading(*args): | |
| return (gr.update(visible=True), *process_input(*args), gr.update(visible=False)) | |
| text_input.change( | |
| fn=process_input, | |
| inputs=[text_input, image_input], | |
| outputs=[alert_html, analysis, output_col] | |
| ) | |
| analyze_button.click( | |
| fn=with_loading, | |
| inputs=[text_input, image_input], | |
| outputs=[loading, alert_html, analysis, output_col] | |
| ) | |
| clear_button.click( | |
| fn=lambda: ("", None, "", "", gr.update(visible=False)), | |
| inputs=[], | |
| outputs=[text_input, image_input, alert_html, analysis, output_col] | |
| ) | |
| # Launch the app | |
| if __name__ == "__main__": | |
| demo.launch() |